fix: Add scalar checks for n and fill_value parameters in shift (…

…#21292)

crates/polars-plan/src/dsl/function_expr/shift_and_fill.rs

@@ -42,20 +42,14 @@ fn shift_and_fill_with_mask(s: &Column, n: i64, fill_value: &Column) -> PolarsRe
 
 pub(super) fn shift_and_fill(args: &[Column]) -> PolarsResult<Column> {
     let s = &args[0];
-    let n_s = &args[1];
-
-    polars_ensure!(
-    n_s.len() == 1,
-    ComputeError: "n must be a single value."
-    );
-    let n_s = n_s.cast(&DataType::Int64)?;
+    let n_s = &args[1].cast(&DataType::Int64)?;
     let n = n_s.i64()?;
 
     if let Some(n) = n.get(0) {
         let logical = s.dtype();
         let physical = s.to_physical_repr();
         let fill_value_s = &args[2];
-        let fill_value = fill_value_s.get(0)?;
+        let fill_value = fill_value_s.get(0).unwrap();
 
         use DataType::*;
         match logical {

crates/polars-plan/src/plans/conversion/functions.rs

@@ -49,6 +49,12 @@ pub(super) fn convert_functions(
 
     let e = to_expr_irs(input, arena)?;
 
+    // Validate inputs.
+    if function == FunctionExpr::ShiftAndFill {
+        polars_ensure!(&e[1].is_scalar(arena), ComputeError: "'n' must be scalar value");
+        polars_ensure!(&e[2].is_scalar(arena), ComputeError: "'fill_value' must be scalar value");
+    }
+
     if state.output_name.is_none() {
         // Handles special case functions like `struct.field`.
         if let Some(name) = function.output_name() {

py-polars/polars/dataframe/frame.py

@@ -9302,8 +9302,8 @@ def shift(self, n: int = 1, *, fill_value: IntoExpr | None = None) -> DataFrame:
             Number of indices to shift forward. If a negative value is passed, values
             are shifted in the opposite direction instead.
         fill_value
-            Fill the resulting null values with this value. Accepts expression input.
-            Non-expression inputs are parsed as literals.
+            Fill the resulting null values with this value. Accepts scalar expression
+            input. Non-expression inputs are parsed as literals.
 
         Notes
         -----

py-polars/polars/expr/expr.py

@@ -2638,7 +2638,7 @@ def shift(
             Number of indices to shift forward. If a negative value is passed, values
             are shifted in the opposite direction instead.
         fill_value
-            Fill the resulting null values with this value.
+            Fill the resulting null values with this scalar value.
 
         Notes
         -----

py-polars/polars/lazyframe/frame.py

@@ -5621,8 +5621,8 @@ def shift(
             Number of indices to shift forward. If a negative value is passed, values
             are shifted in the opposite direction instead.
         fill_value
-            Fill the resulting null values with this value. Accepts expression input.
-            Non-expression inputs are parsed as literals.
+            Fill the resulting null values with this value. Accepts scalar expression
+            input. Non-expression inputs are parsed as literals.
 
         Notes
         -----

py-polars/polars/series/series.py

@@ -5436,8 +5436,8 @@ def shift(self, n: int = 1, *, fill_value: IntoExpr | None = None) -> Series:
             Number of indices to shift forward. If a negative value is passed, values
             are shifted in the opposite direction instead.
         fill_value
-            Fill the resulting null values with this value. Accepts expression input.
-            Non-expression inputs are parsed as literals.
+            Fill the resulting null values with this value. Accepts scalar expression
+            input. Non-expression inputs are parsed as literals.
 
         Notes
         -----

py-polars/tests/unit/operations/test_shift.py

@@ -5,6 +5,7 @@
 import pytest
 
 import polars as pl
+from polars.exceptions import ComputeError
 from polars.testing import assert_frame_equal, assert_series_equal
 
 
@@ -119,3 +120,61 @@ def test_shift_fill_value_group_logicals() -> None:
     result = df.select(pl.col("d").shift(fill_value=pl.col("d").max(), n=-1).over("s"))
 
     assert result.dtypes == [pl.Date]
+
+
+def test_shift_n_null() -> None:
+    df = pl.DataFrame({"a": pl.Series([1, 2, 3], dtype=pl.Int32)})
+    out = df.shift(None)  # type: ignore[arg-type]
+    expected = pl.DataFrame({"a": pl.Series([None, None, None], dtype=pl.Int32)})
+    assert_frame_equal(out, expected)
+
+    out = df.shift(None, fill_value=1)  # type: ignore[arg-type]
+    assert_frame_equal(out, expected)
+
+    out = df.select(pl.col("a").shift(None))  # type: ignore[arg-type]
+    assert_frame_equal(out, expected)
+
+    out = df.select(pl.col("a").shift(None, fill_value=1))  # type: ignore[arg-type]
+    assert_frame_equal(out, expected)
+
+
+def test_shift_n_nonscalar() -> None:
+    df = pl.DataFrame(
+        {
+            "a": [1, 2, 3],
+            "b": [4, 5, 6],
+        }
+    )
+    with pytest.raises(
+        ComputeError,
+        match="'n' must be scalar value",
+    ):
+        # Note: Expressions are not in the signature for `n`, but they work.
+        # We can still verify that n is scalar up-front.
+        df.shift(pl.col("b"), fill_value=1)  # type: ignore[arg-type]
+
+    with pytest.raises(
+        ComputeError,
+        match="'n' must be scalar value",
+    ):
+        df.select(pl.col("a").shift(pl.col("b"), fill_value=1))
+
+
+def test_shift_fill_value_nonscalar() -> None:
+    df = pl.DataFrame(
+        {
+            "a": [1, 2, 3],
+            "b": [4, 5, 6],
+        }
+    )
+    with pytest.raises(
+        ComputeError,
+        match="'fill_value' must be scalar value",
+    ):
+        df.shift(1, fill_value=pl.col("b"))
+
+    with pytest.raises(
+        ComputeError,
+        match="'fill_value' must be scalar value",
+    ):
+        df.select(pl.col("a").shift(1, fill_value=pl.col("b")))